Generative Adversarial Optimization (GAO)


[中文]  [English] [pdf] [code]

  • Tan, Ying, and Bo Shi. “Generative Adversarial Optimization.” International Conference on Swarm Intelligence, 2019, pp. 3–17.

GAO is an novel optimization algorithm inspired by Generative Adversarial Network (GAN) and Fireworks Algorithm (FWA). It utilizes a discriminator and a generator. Usually the distribution of the objective function is complicated. It is very difficult to directly to generate a complete solution from a noise and current solution, which will result in uneven quality of the generated solutions, unstable training procedure and mode collapse problem. The generator generates a generated solution from a noise vector (and optionally adding a current solution). The discriminator is to discriminate whether a generated solution is better than a current solution. The objective function labels a pair of current solutions and generating solutions, then trains the discriminator. The generator's training hopes to maximize the probability that the generated solution is judged to be better than the current solution by the discriminator. Discriminator training is expected to minimize classification errors ... [中文]  [English]


This work has been published as a conference paper at ICSI 2019.


We have open sourced our code implementation for GAN, and we sincerely hope that more people can collabrate with us and further develop this algorithm.

Related Work

  • Tuba, Milan, and T. U. B. A. Eva. "Generative Adversarial Optimization (GOA) for Acute Lymphocytic Leukemia Detection." Studies in Informatics and Control 28.3 (2019): 245-254.